How to Read the Coinbase Premium

Every number in this series so far has come off a derivatives venue. This one is the most-watched spot number in crypto, and it's the simplest: the same coin, priced on two exchanges, and the gap between them. Coinbase is where American retail and American funds buy bitcoin for dollars. Binance is where the rest of the world buys it for Tether. So when Coinbase's price sits above Binance's, the story goes, Americans are paying up — and when it sits below, they're selling.

Both venues publish hourly candles going back years, and they overlap from 17 August 2017, Binance's first day. That's 3,305 days of the two prices side by side, 79,320 hourly readings, for bitcoin and for ether. This article is what they show.

What the number measures

Take one hour. Bitcoin closes at some price on Coinbase's BTC-USD market and at some price on Binance's BTCUSDT market. The premium is the first over the second, minus one, and it's small enough that it's quoted in basis points — hundredths of a percent. A reading of +10 bp means Coinbase was 0.1% above Binance that hour. A day's reading here is the mean of its 24 hours.

Two things about the number before the findings. First, it's not a spread you can trade for free. To collect a positive premium you sell on Coinbase and buy on Binance, then move bitcoin one way and dollars the other; the fees, the withdrawal times and the counterparty risk are the cost, and the premium is what's left after the arbitrageurs have taken everything above that cost. Second — and this is the one the dashboards skip — Binance's price isn't in dollars. It's in Tether. So the raw premium is two gaps in one: how much more Coinbase's buyers pay, and how far Tether is from a dollar that day.

Finding 1: the gap has closed

In 2017 the premium was a real number. The median day was +57 bp, the 95th percentile +518 bp, and inside a typical day the gap moved 183 bp between its high and low hour. On 24 December 2017, with bitcoin at $13,500 on Binance, Coinbase's price averaged 10.0% higher for the whole day. Every one of the five widest days in the record is from September or December 2017.

Then the arbitrage got cheaper and faster, and the band closed. By 2020 the middle 90% of days sat inside −15 bp to +19 bp; by 2023, −7 bp to +11 bp. The intraday range went from 183 bp to 16 bp to 4.5 bp. Ether's series closed on the same schedule: 222 bp of intraday range in 2017, 5.1 bp in 2026. A premium that used to be worth a percent is now worth less than a typical bid-ask spread.

Finding 2: what's left is mostly Tether

Here's the part the index hides. Binance quotes bitcoin in USDT, and USDT is not always a dollar. When Tether trades at $1.0004, every Binance price is 4 bp too low in dollar terms, and the raw premium reads +4 bp with nobody on Coinbase paying up for anything. When Tether trades at $0.9995, the premium reads −5 bp and no American is selling.

Adjust for Tether and the premium mostly disappears. Bitcoin's raw median was +4.4 bp in 2020 and +4.4 bp in 2021; Tether traded at $1.0004 and $1.0004; the adjusted medians are +0.5 bp and +0.7 bp. This year the raw median is −6.0 bp — a "Coinbase discount" that has lasted all year, 83% of days negative — and Tether's median price is $0.9995. Adjusted, the discount is −0.4 bp. Since 2020 the adjusted premium's typical size has never exceeded 2.6 bp in any year. Ether's is the same: 0.9 bp this year.

The same thing explains the widest days in both directions. The narrowest reading in the whole record is 15 October 2018: −567 bp, Coinbase 5.7% below Binance, with the next four narrowest days the four that followed. That's not American selling — it's the day Tether lost its peg and traded as low as $0.92. Bitcoin priced in a Tether worth less than a dollar jumped on Binance while its dollar price on Coinbase sat still, and the "discount" was the market marking Tether down. The day USDC broke its peg, 11 March 2023, ran the other way: a raw premium of +80 bp, Tether at $1.0077 on Coinbase as money fled USDC into it, and an adjusted premium of +3.1 bp.

Read down the adjusted column and the days that were supposed to be about American demand are quiet. The ETFs' first day, 11 January 2024: +0.6 bp. The largest cascade in the record, 10 October 2025: −11.6 bp, with an intraday range of 37 bp. The widest adjusted days since 2020 are all plumbing days. The morning after Black Thursday, 13 March 2020: +65 bp adjusted, with the gap swinging 582 bp inside the day, as exchanges buckled under the load and the arbitrage that normally closes the gap stopped running. Terra's collapse, 12 May 2022: −59 bp, on a day Tether itself wobbled to $0.998 and a daily close is a poor measure of its price. The gap opens when the pipes are blocked, not when America buys.

Finding 3: it has no session

If the premium were American buying, you'd expect it to switch on when America wakes up: a bigger gap from 9am to 5pm New York time, a smaller one while Asia trades. So split every day's hourly readings into three sessions and take each year's median.

They're the same number. Since 2018 no year's three sessions have differed by more than 2.9 bp, and since 2021 by more than 0.6 bp. The weekend isn't different either: +1.1 bp on a weekday, +2.0 bp on a weekend, raw; +0.1 bp and +0.5 bp adjusted. The premium doesn't turn on at the New York open, because the arbitrage runs around the clock and carries whatever gap exists across every session.

What the premium is, instead, is slow. Today's reading is highly correlated with yesterday's (0.84), still correlated with the reading a week ago (0.53) and faintly with a month ago (0.22). It's a level that drifts — which, once you know that most of it is Tether's dollar price, is exactly what you'd expect. A stablecoin's deviation from par is a slow thing.

And it barely responds to the day's move. On a day bitcoin rises 5% or more, the raw premium's median change is +1.3 bp; on a day it falls 5% or more, −0.2 bp; on a flat day, −0.1 bp. Adjusted, the 5% up day moves the premium −0.7 bp and the 5% down day +0.1 bp. The correlation between the day's return and the day's change in the premium is -0.07. Whatever a big US up day looks like, it doesn't look like Coinbase's price pulling away from Binance's.

Finding 4: a fingerprint of the month that just happened

The most common use of the index is as a signal: a rising premium means the US is bidding and the move has legs; a falling one means the bid has gone. Take every day since 2020, measure its raw premium against the median of the previous 90 days, split the days into five bands, and look both ways — a month back and a month forward.

Read from the left, the table is a staircase: the bottom band came after a −0.9% month, the top band after +14.6%. A high premium is what the end of a rally looks like — and given finding 2, what it mostly means is that Tether trades a few basis points above par when people are buying crypto with it. Read from the right, the staircase is gone. The next 30 days after the top band averaged +3.8%, up 54% of the time; after the bottom band, +0.3%, up 47% of the time. The bands in between don't line up in order.

Split the same test by year and the forward edge changes sign. The top decile beat the bottom decile over the next month in 5 of 7 years, and the sign of the gap flipped 4 times between consecutive years. On the adjusted premium — the part that's actually Coinbase against Binance — the top decile since 2 August 2021 was followed by +1.4% over the next month and the bottom decile by +1.9%, which is nothing.

The one place the premium ever did forecast is the era it was already a percent wide. Across the whole record from 2017, the bottom band — readings 15 bp or more below the 90-day median — was followed by −2.1% over the next month, up 42% of the time. 141 of those 318 days are from 2018, the year of the Tether depeg and the November crash that followed it. That's not a signal you can use. It's one bear market, read through a broken peg.

Finding 5: the history is a history of who was trading

One more column from the year table deserves its own paragraph. In 2017, Coinbase carried 79% of the two venues' combined dollar volume in bitcoin; Binance was a month old. By 2023 Coinbase's share was 13%. In 2026 it's 32% — the American venue's share more than doubled after the spot ETFs started trading in January 2024 and their authorised participants started sourcing bitcoin in dollars.

That's the honest version of the index's story. American demand shows up in this data — as volume, as the share of the market that trades in dollars. What it doesn't show up as is a price gap, because a price gap is what arbitrage exists to remove, and by 2020 it was removing it to within a few basis points around the clock. The premium of 2017 was a real thing: a retail venue carrying 49% of the two venues' volume on Christmas Eve, priced 10% above the rest of the world. The premium of 2026 is Tether's peg, plus noise.

How to read it yourself

• Check Tether first. Before reading any premium against a USDT venue, look up USDT-USD. If Tether is at $1.0005, a +5 bp premium is nothing; if it's at $0.9992, a −8 bp "discount" is nothing. Since 15 December 2018 the middle 90% of adjusted days sit between −6 bp and +8 bp.
• Expect basis points, not percent. The middle 90% of bitcoin's days since 2020 sit within a few basis points of zero. A reading of ±1% is a broken peg, a halted withdrawal, or 2017.
• Don't wait for the New York open. The three sessions print the same median; the gap you see at 9am is the gap you'd have seen at 3am.
• A high reading tells you the last month was good. The top decile came after +14.6%. It didn't tell you about the next month, and on the adjusted series it told you nothing at all.
• The widest days are plumbing, not demand. The five narrowest days in the record are Tether's depeg; the widest adjusted days since 2020 are the morning after Black Thursday and Terra's collapse. Read a spike as a question about the pipes before a question about buyers.
• If you want American demand, look at volume share. Coinbase's share of the pair's dollar volume went from 13% in 2023 to 32% in 2026. That's the ETF, and it's visible. The price gap isn't.

Method, and what this cannot tell you

Hourly closes for BTC-USD and ETH-USD from Coinbase Exchange's public candles and for BTCUSDT and ETHUSDT from Binance's public spot klines, for every UTC hour from 17 August 2017, the first day both venues traded, to 17 September 2026. The raw premium is the Coinbase close over the Binance close minus one, in basis points; a day's reading is the mean of its hourly readings, and a day needs at least 20 matched hours to count. The adjusted premium divides Binance's Tether price by Tether's dollar price first: Coinbase's own USDT-USD daily close from 4 May 2021, and before that, from 15 December 2018, Binance's USDC-USDT daily close inverted. Session medians are of the raw hourly readings in the US (13:00-21:00 UTC), Asian (00:00-08:00) and European (08:00-13:00) windows. The forward tests measure each day's premium against the median of the previous 90 days, split the days into bands at the 10th, 30th, 70th and 90th percentiles, and report the mean spot return over the 30 days before and the 7 and 30 days after, using Binance's close. Coinbase's share of volume is Coinbase's hourly base volume times its close, over that plus Binance's quote volume.

• Two venues only. Coinbase against Binance is the pair the popular index uses, and it is the pair we measured; a premium against Kraken, Bitstamp or the ETF creation price is a different number.
• The adjustment assumes a stablecoin is a dollar. From May 2021 we use Coinbase's own USDT-USD market, which is a real dollar price. Before that we infer Tether's price from USDC-USDT on Binance, which is only right when USDC itself trades at par — it did not on 11 March 2023, which is why that day is read from the Coinbase series, and it may not have on other days in 2019-2021.
• No adjustment before December 2018. The 2017 and 2018 figures are raw, so Tether's depeg in October 2018 and its dollar price during the 2017 mania are inside them, and named as such.
• Close against close, once an hour. A gap that opens and closes inside an hour is invisible here; the intraday range is the range of hourly readings, not of every trade.
• The forward bands overlap heavily and a year of one regime dominates several of them — the top band's most common year is named in each table for that reason.
• Coinbase volume is base volume times the hourly close, an approximation of dollar volume; Binance's is the kline's quote volume.